The comprehension of severe criticality accident is a key issue in Gen-IV neutronics and safety. Within the future zero-power experimental physics reactor (ZEPHYR), to be built in Cadarache in the next decade, innovative approaches to reproduce high temperature partially degraded Gen-IV cores into a critical facility is being investigated. This work presents the first attempt to represent a fuel assembly of sodium-cooled fast reactor severe criticality accident based on surrogate models. One identified way to construct such representative configuration is to use MASURCA plates stockpile (MOX, UOx, Na, U, and Pu metal) in a fast/thermal coupled core to model a stratified molten assembly. The present study is the first step in a more global approach to full core analysis. The approach is based on a nature-inspired metaheuristic algorithm, the particle swarm optimization algorithm, to find relevant ZEPHYR configuration at 20°C that exhibits characteristics of (2000-3000°C) molten MOX assembly in a stratified metal arrangement in a reference sodium-cooled fast reactor core. Thus, the underlying research question of this study is whether it is possible to represent temperature-related reactivity effects occurring at fuel meltdown temperatures in a power reactor as density-related reactivity effects at the operation temperature of a zero-power reactor, and if so, how should it be done? The calculations are based on a Serpent-2 Monte Carlo sensitivity and representativity analysis using the Cadarache's cross sections covariance data (COMAC). The single fuel assembly studies show that it is possible to represent the multiplication factor with a representativity factor greater than 0.98. As for reactivity variations, it is possible to achieve a satisfactory representativity factor of above 0.85 in all the presented cases. The representativity process demonstrates that temperature effects could be translated into density effects with good confidence. A complementary analysis on modified nuclear data covariance matrix demonstrates the importance of selecting consistent and robust uncertainties in the particle swarm optimization algorithm. This work provides insights on the behavior of the representativity scheme in different core states and shades some light on the problem in hand.